📊 What’s new?
Unlike previous ML models treating months independently, the LSTM processes the full monthly climate sequence and carries information forward in time, allowing explicit representation of cumulative seasonal processes such as snow accumulation and progressive melt.
🚨 New preprint 🚨
We present the first application of an LSTM to glacier mass balance modeling, within the Mass Balance Machine (MBM). With its recurrent memory, the model captures cumulative seasonal processes and generalizes across Swiss glaciers 🏔️❄️
@vaw-glaciology.bsky.social
(3) Why it matters:
🌊 Glaciers = key freshwater reservoirs & climate indicators
🧊 Few are directly monitored
🤖 ML (XGBoost) helps bridge data gaps
Mass Balance Machine on GitHub 👉 lnkd.in/eyFB5SZp
Link to paper 👉 lnkd.in/eD-H5DT8
#Glaciology #MachineLearning #Cryosphere #ClimateChange #Hydrology
(2) Tested on independent Norwegian glaciers, MBM was compared with GloGEM, OGGM & PyGEM:
✅ RMSE ≈ 0.59–1.00 m w.e.
✅ Bias: –0.01 to +0.04 m w.e.
✅ Outperforms conventional models for seasonal MB
✅ Promising transferability across glaciers & climates 🌍
🚨 Introducing the Mass Balance Machine (MBM): a #data-driven glacier MB model using topographic + climate features 🏔️ 👉 "Machine learning improves seasonal mass balance prediction for unmonitored glaciers” (Sjursen et al., 2025) doi.org/10.5194/tc-1...
@vaw-glaciology.bsky.social
Greetings from Marijn van der Meer at the #Ellis Summer School in Jena 🇩🇪! She presented her work on the #MassBalanceMachine. The school brings together AI & climate science & is co-organised by top institutes & supported by @climatechangeai.bsky.social, @esa.int Academy & more.
🧊 New preprint:
We present the Mass Balance Machine (MBM), an XGBoost-based model predicting glacier mass balance at high resolution, even for glaciers without in situ data.
Applied to Norwegian glaciers, MBM generalizes well, outperforming TI models in seasonal mass balance prediction.
(3/3) Using just two predictors obtained through dimensionality reduction techniques, miniML-MB can closely match the PMB for individual glacier sites, surpassing the PDD model for most sites as long as predictions are made within a range of meteorological conditions similar to the training set.
(2/3) Our dimensionality reduction framework singles out summer temps (May–Aug) & winter precip (Oct–Feb) as key drivers of glacier PMB. Unlike PDD models, our ML approach directly selects predictors from data—boosting performance & validating climatic drivers in Swiss glaciers.
(1/3) miniML-MB is designed to simulate annual point mass balance at individual glacier sites using meteorological variables (air temperature and total precipitation). We rely on data collected at 28 individual measurement sites across the Swiss Alps:
🚨Introducing miniML-MB: a #MachineLearning model using XGBoost to estimate glacier mass balance from very small datasets! Applied in the Swiss Alps, it pinpoints key drivers—May–Aug temp & Oct–Feb precip—and outperforms a basic PDD model. @vaw-glaciology.bsky.social
tc.copernicus.org/articles/19/...
Greetings from sunny, winter-cold Oslo and the Global Glacier Modelling Workshop 2025! ❄️ Several #VAW members joined to present exciting new developments on the Global Glacier Evolution Model as well as the Mass Balance Machine. Three productive days of great discussions! 💻🇳🇴
In case of interest in the Mass Balance Machine, check out our GitHub 🧊 github.com/ODINN-SciML/...
🚨 MSc thesis opportunity at @vaw-glaciology.bsky.social The thesis focuses on applying a glacier mass balance model based on machine learning, driven by climate variables and topographical features. This is a great opportunity to apply data science to a real-world problem :) 🏔️
